HydraIssues

Reflector fleet: one probe reflector VM per candidate cloud region
open unclassified Project: hydracluster Parent: #695 Reporter: 14 Sep 2026 10:16

Description

Child of #695 (plan Phase 2). RULING 2026-09-14: hyperscalers only, see the ruling block in the #695 plan.

Provision one cheap CPU VM (3 to 5 EUR/month) per candidate region after the Phase 1 ping prune. Candidate regions are now exactly four: azure-westeurope, aws-eu-central-1, aws-eu-west-3, gcp-europe-west1. No zero-egress reflectors (LeaderGPU, OVH, Hetzner are out of scope per the ruling; #710 closed).

Each reflector runs hydraprobe server (systemd, HMAC, UDP 2112) and a WireGuard peer. Peering stage 1: all reflectors peer at the Brussels hub. Stage 2: direct neck peers for the top 2 regions per venue. Add reflectors to the #42 proximity poller as pseudo-districts. After gate G2, delete losing reflectors; the winner's reflector stays as the district monitoring peer.

Budget: 12 to 20 EUR total for about 1 month. Related: #42.